An Automatic Labeling System for Broadcast News Based on Recognition and Dynamic Programming

摘要

Large amount of precise labeled corpus is the basic of speech recognition, and labeling is a time-consuming and labor-consuming work. It's easy to get the audios and texts of the Broadcast News, but still requires a great deal of manual labeling. In this paper we build an automatic labeling system for Chinese broadcast news. By using the texts downloaded from internet, through three steps as audio segmentation, speech recognition, sentence-level alignment based on dynamic programming, a quasi-precise labeled document in sen-tence is generated.